Project 22: Custom Linear Layer Builder

ML Engineer

Builds on these lessons

Current Task

Objective

Overriding build() lets a custom layer create its weights lazily, once it knows the actual input shape — the low-level pattern behind every built-in Keras layer.

Task: write a Linear custom layer that creates its own weight in build(), then apply it to a batch of inputs.

index.py

import tensorflow as tf
from tensorflow.keras import layers

class Linear(layers.Layer):
    def build(self, input_shape):
        self.w = self.add_weight(shape=(input_shape[-1], 4), initializer='random_normal')

    def call(self, inputs):
        return tf.matmul(inputs, self.w)

layer = Linear()
output = layer(tf.ones((2, 3)))
print(output.shape)

* Hint: Correct characters turn green, incorrect ones turn red.

Live Preview

🖼️

Verify your code to see the preview

← Previous Next →